Machine learning for business applications
The implementation of machine learning in business applications.
Recognizing handwritten text has been among the earliest use cases for deep learning. State-of-the-art technology can solve this task as well as humans – or even better. However, to take full advantage of these new, time-saving possibilities, they need to be seamlessly integrated with today’s core business technologies. Read on to find out how we effortlessly integrated a neural network into an Eclipse Scout business application.
In a proof of concept study, we chose a very common use case: handwritten payment slips that must be digitized and transferred into a core banking system. We used our open source framework Eclipse Scout to build the payment slip reader’s user interface. To build, train and use the neural network that recognizes handwritten digits, we used Deeplearning4j, the most comprehensive and mature deep learning Java library currently available. Because it is a Java library, the integration of Deeplearing4j with the Eclipse Scout framework proved to be as smooth as we expected.
Training our neural network
After implementing the Scout application’s user interface, we were able to integrate a suitable neural network model in the code. Now we needed to train the network; to do so, we asked people to write numbers in their everyday hand writing style and used this data to train our neural network model.
“The neural network can be easily integrated into Eclipse Scout.”
Christoph Bräunlich, software engineer at BSI
Integration proved to be straightforward
We demonstrated how simple it is to integrate a neural network into a business application with our proof of concept. As the described integration is domain-independent, it can be applied to all types of business applications, such as BSI CRM or any client project. Payment slips might soon be a thing of the past, but the role of machine learning is expected to grow rapidly across all major industries. There are already many use cases for machine learning today: text analysis (e.g. sentiment analysis or automatic assignment of communications to processes or agents) and image processing (e.g. recognition of handwritten or machine documents, photography manipulation) are only some examples that will keep us occupied in the future.
Want to learn more or try it yourself?
Check out our Scout Blog for a more detailed technical description of our use case. There, we also share the full source code of the demo application, “Anagnostes,” on GitHub.